A hybrid fuzzy–deep learning framework for real-time cyber-attack detection in smart energy grids

The growing adoption of communication and automation technology in smart energy grids has greatly amplified their vulnerability to cyber-attacks, and therefore, intrusion detection in a timely and reliable manner has become a pressing necessity. In this context, this paper will seek to fill this gap by introducing a hybrid fuzzy-deep learning framework, which combines uncertainty-sensitive feature modeling with spatiotemporal deep representation learning. A convolutional recurrent neural network is used to encode coordinated spatial and temporal patterns of attacks in the form of graded representations of raw grid and network measurements using the help of fuzzy logic. With this integration it is possible to have end to end learning in nonstationary and ambiguous operating conditions. The proposed framework achieves a detection accuracy of 97.6%, an F1-score of 0.96, and a false positive rate of 3.1%, outperforming representative machine learning and deep learning baselines. In addition, the average detection latency of 29.6 ms confirms its suitability for real-time monitoring applications. The primary value of the work is its ability to show that the systematic combination of the uncertainty modeling based on fuzzy with deep spatiotemporal learning can considerably increase the reliability of detection and the feasibility of operation, which can provide a viable way to achieve the goal of resilience in cybersecurity of smart energy grids.